模糊特征下基于神经网络的工程图纸标注符号识别研究

On Recognizing Symbols Marked on Engineering Drawings

提出了一种基于人工神经元网络对机械工程图纸标注符号识别的方法，并对基于特征输入和基于点阵输入两种神经网络分类器的特点进行了比较．研究结果表明，在模糊特征下，神经网络方法对工程图纸标注符号的智能识别完全能达到实用化要求．

Automatic input and intelligent recognition of marked symbols on engineering drawings is one key problem still remaining to be solved in Chinese CAD research. We apply the rapidly emerging neural network technology to its solution.We use BP algorithm and compare the merits of two different input methods: raster input and fuzzy feature input' After fairly full discussion of the results given in Table 2, wefind that fuzzy feature input method is better.Recognition rates are given in Table 4. Two test sample sets are used: test 1 (1980numbers scanned from drawings that range from 0 to 9) and test 2(170 symbols ). Trainingsample set B includes not only set A but also those samples in set A incorrectly recognized.In our opinion, recognition rates given in Table 4 can be said to satisfy fallly well the practical needs for converting symbols marked on engineeTing drawing to 2D CAD structure.